What determines the FPS in a cell?

[sirherrbatka]

I have no idea how valid this is but gus used to state that our main problem is the number of objects that needs to be rendered (calls to GPU). Because of this batching looks like a promising method of improving rendering speed but it's worth to mention that morrowind appears to be a tricky game for batching.

[nopoe]

Potentially worth bringing up are those builds that load 11x11 cells and render them all with no fog. I did one and it didn't even touch my GPU but my CPU was maxed out. I would normally say this is physics or scripting, but when I looked up, causing the game to only render a skybox, the framerate went up dramatically. I don't remember the exact numbers but I think it was around 2fps around vivec city to around 20fps looking up. There is obviously no substitute for proper profiling, but if I had to guess I'd say that rendering the city caused an order-of-magnitude drop in performance on the CPU side but essentially no load on the GPU side.

To be clear, I'm perfectly aware that this is an edge case and should not be considered for optimization for until way after 1.0. I simply think it might be useful information to have.

[sirherrbatka]

I'm not sure but it seems to be a logical to me. There is a plenty of objects with different textures. GPU first needs to get information from CPU about objects that needs to be rendered so most of time it does not render but wait for CPU instead.

I'm not a programmer at all and I have no idea how does it work. This is just my guess.

[Lazaroth]

So, here's an interesting observation;

My Nvidia 460 GTX just crapped out on me. The card I have as a replacement until I get my new card is an old passive AMD HD 2600. According to a GPU comparison site, it has about 1/8 of the performance, not counting that my 460 was factory overclocked.

The interesting part is that my FPS with the same maxed settings is not that much lower. E.g. Wilderness 60->40 and Balmora 30->20. So yeah, the CPU is definitively much more important.

[Tarius]

So, here's an interesting observation;

My Nvidia 460 GTX just crapped out on me. The card I have as a replacement until I get my new card is an old passive AMD HD 2600. According to a GPU comparison site, it has about 1/8 of the performance, not counting that my 460 was factory overclocked.

The interesting part is that my FPS with the same maxed settings is not that much lower. E.g. Wilderness 60->40 and Balmora 30->20. So yeah, the CPU is definitively much more important.

Well yes, for MW, CPU is more important as they ade MW rely on it more than GPU. I suppose that GPU was and in many cases still is used for physics mostly(depending on what the program is). Of course it would seem things are moving towards using the GPU for more than just physics today of course. Since MW wasnt physics heavy, and draw distance for everything wasnt very far, they probably didnt see a need to utilize the GPU much.
Nothing excels at raw calculations more than a GPU though. This is why they are starting to utilize them to build cheap super computers.

[claudekennilol]

I guess they are planning on doing it later on (closer to 1.0) which sounds a bit strange to me but hey, they know what they are doing.

"We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil" -- Knuth

Premature optimization: Coding early-on for perceived efficiency, sacrificing good design, maintainability, and sometimes even real-world efficiency

Please read this:
http://en.wikipedia.org/wiki/Optimizati ... o_optimize

Thank you for linking those. I was thinking that exact same quote when reading the post you responded this to and was about to go off and find something along the lines of your wikipedia entry that I could link to to support it. Glad there're quality programmers working on this project ^_^.

[wheybags]

It seems like you are referring to the location where your Hlaalu Stronghold would be if you complete their questline (something plateau). That is a heavily scripted area, but everything is disabled at first, so unless OpenMW isn't properly handling disabled items then I can't see any other reason for lag.

Actually, scripts do run on disabled items, in both vanilla and openmw.

[TorbenC]

It seems like you are referring to the location where your Hlaalu Stronghold would be if you complete their questline (something plateau). That is a heavily scripted area, but everything is disabled at first, so unless OpenMW isn't properly handling disabled items then I can't see any other reason for lag.

Actually, scripts do run on disabled items, in both vanilla and openmw.

Interesting, well then it is just a matter of waiting until OpenMW is at the point where you guys can start optimizing stuff.

[Tarius]

It seems like you are referring to the location where your Hlaalu Stronghold would be if you complete their questline (something plateau). That is a heavily scripted area, but everything is disabled at first, so unless OpenMW isn't properly handling disabled items then I can't see any other reason for lag.

Actually, scripts do run on disabled items, in both vanilla and openmw.

Interesting, well then it is just a matter of waiting until OpenMW is at the point where you guys can start optimizing stuff.

Zini says it likely wont be until right before 1.0 when things start getting optimized. There are varying thoughts on the matter from what zini says to they should optimize some along the way though.
Considering it is likely that 1.0 will be here no later than the beginning of next year, it wont be terribly much longer.

[TorbenC]

Zini says it likely wont be until right before 1.0 when things start getting optimized. There are varying thoughts on the matter from what zini says to they should optimize some along the way though.
Considering it is likely that 1.0 will be here no later than the beginning of next year, it wont be terribly much longer.

Well from what I read of that article Zini is right on target with how a project should be optimized, minor bits along the way and a general optimization near the end.